Pritha Guha
Verified Expert in Engineering
Cloud Architect and Software Developer
Kolkata, West Bengal, India
Toptal member since March 4, 2026
Pritha is a senior MLOps, SRE, and cloud platform engineer with 11+ years building production infrastructure on AWS, GCP, and Kubernetes. Experienced in fast-paced SaaS startups and large enterprises, she owned end-to-end infrastructure, including MLOps stack (Kubeflow, KubeRay, KServe, GPU, RAG pipelines, agentic AI, Amazon Bedrock, Gemini). She’s cut costs by 60% and Kubernetes upgrade downtime by 97%, and increased PostgreSQL throughput by 40%. She's led SOC 2 Type II compliance end-to-end.
Portfolio
Experience
- Kubernetes - 8 years
- Terraform - 8 years
- Docker - 8 years
- Amazon Web Services (AWS) - 8 years
- Google Cloud Platform (GCP) - 7 years
- Helm - 6 years
- Agentic AI - 2 years
- Kubeflow - 1 year
Preferred Environment
Amazon Web Services (AWS), Google Cloud Platform (GCP), Kubernetes, Agentic AI, Machine Learning Operations (MLOps), Site Reliability Engineering (SRE), Platform Engineering
The most amazing...
...success has been reworking the architecture of an app hosted on GCP and Kubernetes, leading to a 63% reduction in costs and a 97% reduction in upgrade downtime.
Work Experience
AI/ML Infrastructure Engineer | MLOps
Turbot
- Delivered a production anomaly detection platform on AWS EKS, using Kubeflow Pipelines for orchestration, KServe for model serving, IsolationForest on cloud logs, and Katib for hyperparameter optimization. Built the proof of concept using code generation tools and designed the full production infrastructure.
- Integrated a LangGraph agent with the anomaly detection platform that intercepts flagged events, retrieves changes since the model's last training run via RAG, and feeds that context to an LLM on Amazon Bedrock to determine if it is a genuine bug.
- Delivered a distributed machine learning platform on GKE using Ray, KubeRay, and Vertex AI for GPU-based training, applied to tag resolution (entity resolution), with keyless authentication via Workload Identity and Spot instance fault tolerance.
- Built an internal tool using a RAG pipeline and MCP server, mapping natural language queries to 200+ Flowpipe pipelines via pgvector embeddings, PostgreSQL storage, and integrating Amazon Bedrock to translate user intent into infrastructure actions.
- Developed internal agentic AI systems for FinOps and Operations using CrewAI, Google ADK, and Amazon Bedrock AgentCore multi-agent workflows, orchestrating autonomous task planning, tool use, and cross-agent handoffs.
- Managed GPU infrastructure for AI/ML workloads across GKE and EKS, including H100 and A100 node pools with CUDA workload configuration, MIG partitioning and time-slicing, and autoscaling tuned for variable LLM inference demand.
- Worked on an AI assistant, designed the LiteLLM proxy, exposing a unified OpenAI-compatible API across multiple model providers, with user-selectable model routing.
- Built and maintained Prometheus and Grafana dashboards for GPU utilization, model serving latency, token cost, and pipeline throughput. Implemented OpenTelemetry tracing for full agent decision-chain visibility, along with drift detection alerting, prediction distribution shift monitoring, and HITL escalation rate tracking.
- Worked on prompt injection controls, MCP server governance, Denial of Wallet prevention, RAG pipeline poisoning defense, Zero Trust IAM with keyless Bedrock and Vertex AI access, AI-BOM, and runtime anomaly detection on agent behavior.
Senior Platform Engineer | SRE | Cloud Architect
Turbot
- Architected six production multi-tenant Kubernetes clusters (EKS and GKE), managing full cluster lifecycle. Reduced upgrade downtime by 97% (45 minutes to 15 seconds) via architectural restructuring, PDB configuration, and blue-green deployment.
- Provisioned 200+ AWS resources across commercial and GovCloud environments using Terraform modules and CloudFormation nested stacks, including EC2, VPC, RDS, ECS, S3, IAM, KMS, ElastiCache, SQS, SNS, ALB, and WAF. Set up landing zones and organization-level governance using SCPs and Control Tower.
- Provisioned and maintained 150+ GCP resources using Terraform, including GKE, Compute Engine, Cloud SQL, Cloud Storage, Cloud Run, VPC, IAM, and Artifact Registry; set up landing zones and organization-level governance via Org Policies.
- Built Helm charts for StatefulSets, Deployments, Jobs, Services, NetworkPolicies, Roles, Namespaces, HPA, Secrets, and certificates. Set up PDBs, tuned resource requests and limits, configured quotas, and patched third-party Helm charts using Kustomize.
- Designed secure, lightweight Docker images using Alpine and Debian. Built CI/CD pipelines with GitHub Actions, CodePipeline, and ArgoCD, implementing supply chain security through SBOM generation, Cosign image signing, Trivy and tfsec scanning, and OPA/Rego policy enforcement as deployment gates.
- Cut GCP cost by 63% and AWS by 34% through right-sizing, savings plan, and architectural restructuring. Authored Ansible playbooks and Jenkins (Groovy) pipelines for smoke test automation, validating post-deployment state across environments.
- Handled observability: Prometheus, Grafana, ELK, OpenSearch, OpenTelemetry, CloudWatch, and GCP monitoring, reducing incident detection cadence from eight hours to 20 minutes. Implemented SLI/SLO and error budgeting across all production services.
- Architected zero-trust networking: network policies, mTLS (Istio), SSL (ACM), and HashiCorp Vault. Led SOC2 Type II compliance and redesigned architecture for HIPAA compliance. Led incident response, blameless postmortems, and severity management.
- Reduced PostgreSQL upgrade and migration downtime from eight hours to 50 seconds. Improved throughput by 40% by implementing PgBouncer. Managed Redis across ElastiCache and Kubernetes environments. Implemented SQS, SNS, and Kafka-based event streaming solutions on AWS and within Kubernetes.
- Handled disaster recovery and designed cross-organization, offsite DR for AWS and GCP, including cross-account snapshot sharing, KMS re-encryption in a backup account, HashiCorp Vault backups, and a pilot-light plan for managed service customers.
Associate
Cognizant
- Developed and maintained Python-based production applications supporting large-scale enterprise ETL workflows, improving data processing reliability and automation.
- Automated AWS infrastructure provisioning using Terraform and the AWS SDK, reducing manual configuration effort and improving environment consistency.
- Guided back-end QA validation for enterprise ETL systems, identifying and resolving critical data integrity issues before production release.
- Documented technical workflows, deployment processes, and validation standards, improving team onboarding and operational continuity.
- Collaborated with cross-functional teams to troubleshoot production issues, reducing turnaround time for high-priority defects.
- Contributed to on-prem to AWS migration, decomposing monolithic applications into ECS/Lambda-based microservices.
Experience
Entity Resolution using KubeRay
https://github.com/pritha-athena93/entity-resolution-kuberayData consistency and versioning are managed through Apache Iceberg integrated with Nessie, providing a robust data lakehouse architecture for transactional writes and "git-like" catalog management. The pipeline streamlines the ingestion and processing of large datasets by leveraging GPU Computing for accelerated ML training. By combining Vertex AI for model management and Python-based workflows, the project delivers a production-ready solution for complex data matching and deduplication at scale.
Multi-cloud Golden Path (AWS, GCP, Kubernetes)
https://github.com/pritha-athena93/multi-cloud-golden-pathAnomaly Detection Platform Using Kubeflow and LangGraph Agent
https://github.com/pritha-athena93/anomaly-detectorI integrated a LangGraph agentic layer over the serving endpoint—on every flagged anomaly, the agent retrieves infrastructure changes made since the model's last training run via RAG, feeds that context to an LLM on Amazon Bedrock, and determines whether the anomaly is a genuine bug or an expected consequence of a known change. It reduces false positive triage burden on the ops team without requiring model retraining for every infrastructure change.
Security Challenge Game
https://github.com/pritha-athena93/tfsec-toolI designed the game to gamify infrastructure security concepts and help engineers understand real-world cloud security risks in a hands-on environment. I also implemented automated validation logic to evaluate user fixes and provide feedback based on security rule detection.
Education
Executive Management Postgraduate Diploma in Business Analytics
Institute of Management Technology, Ghaziabad - Online
Bachelor's Degree in Computer Science
St. Xavier's College - Kolkata, India
Certifications
AWS Certified Solutions Architect – Associate
Amazon Web Services
Skills
Libraries/APIs
vLLM, Node.js, AWS Amplify
Tools
Terraform, Grafana, Helm, AWS IAM, AWS CloudFormation, Amazon CloudWatch, Observability Tools, Google Kubernetes Engine (GKE), Amazon EKS, AWS Step Functions, AWS Fargate, GitLab CI/CD, Jenkins, Amazon Elastic Container Service (ECS), Amazon OpenSearch, AWS ELB, Amazon ElastiCache, Amazon Virtual Private Cloud (VPC), Kubernetes HorizontalPodAutoscaler (HPA), GitHub, AWS CodeBuild, Amazon Simple Notification Service (SNS), Amazon Elastic Container Registry (ECR), Apache Airflow, Amazon SageMaker, Istio, OpenTofu, Amazon Cognito, Cloud Development Kit for Terraform (CDKTF), Ansible, AWS CodeCommit, ELK (Elastic Stack), HashiCorp Vault, NGINX, Apache Iceberg, GitLab, DeepSeek
Languages
SQL, Python, Bash, Go, JavaScript, TypeScript, Python 3
Paradigms
DevOps, Event-driven Architecture, Continuous Delivery (CD), Continuous Integration (CI), Role-based Access Control (RBAC), DevSecOps, HIPAA Compliance, Automation, ETL
Platforms
Amazon Web Services (AWS), Google Cloud Platform (GCP), Kubernetes, Docker, Amazon EC2, AWS Lambda, Linux, Kubeflow, Cloud Run, KServe, Apache Kafka, Rancher, NVIDIA CUDA, Vertex AI, CrewAI
Storage
PostgreSQL, Elasticsearch, Amazon S3 (AWS S3), Amazon DynamoDB, MySQL, MongoDB, Redis, On-premise
Frameworks
LangGraph, Ray
Other
Software Design, Networks, Cloud Infrastructure, Prometheus, Identity & Access Management (IAM), Virtual Private Cloud (VPC), Networking, CI/CD Pipelines, Infrastructure as Code (IaC), Containerization, AWS X-Ray, Amazon RDS, SOC 2, Distributed Systems, AWS Auto Scaling, AWS Cloud Security, Cloud Architecture, AWS Cloud Architecture, Multi-tenant Architecture, Multi-tenant SaaS, Site Reliability Engineering (SRE), GitHub Actions, Infrastructure Automation, Cloud, GitOps, Vibe Coding, AWS DevOps, AWS Certified Solution Architect, AWS Cloud Operations, IT Security, Zero Trust, Zero Trust Network Access (ZTNA), Observability, Containers, Compliance, Incident Response, Solution Architecture, Performance, Security, AWS ECS Fargate, Amazon Glacier, Agile DevOps, Infrastructure, SOC Compliance, Monitoring, OpenTelemetry, Architecture, Vulnerability Management, Vulnerability Scanning, API Gateways, Artificial Intelligence (AI), Large Language Models (LLMs), Service Meshes, HIPAA, GPU Computing, MLflow, Cloud Security, Amazon Redshift, Server Optimization, Data Engineering, Agentic AI, APIs, General Data Protection Regulation (GDPR), Agentic AI Systems, AI Automation, Large Language Model Operations (LLMOps), Linux Network Stack, Amazon Bedrock AgentCore, AI Agents, AWS IAM Identity Center, Temporal, Argo CD, System Architecture Design, High Availability (HA), High Availability Disaster Recovery (HADR), Relational Database Services (RDS), Cost Reduction & Optimization (Cost-down), Quality Assurance (QA), Software Development, Machine Learning Operations (MLOps), Self-hosted, KubeVirt, Gemini, Data Science, Business Analytics, Platform Engineering, Hyperparameter Optimization, RAG Systems, Healthcare Software, Amazon Machine Learning, Machine Learning, Telemetry, Distributed Tracing
How to Work with Toptal
Toptal matches you directly with global industry experts from our network in hours—not weeks or months.
Share your needs
Choose your talent
Start your risk-free talent trial
Top talent is in high demand.
Start hiring